基于改进随机游走算法的阴影与遮挡处理方法

Method for Shadow and Occlusion Based on the Improved Random Walk Algorithm

将交互式分割算法与Kalman滤波器结合,提出基于Kalman滤波器的随机游走算法,并将其用于解决交通视频监控中的阴影与遮挡问题.首先利用Kalman滤波器的预测信息对随机游走的计算区域进行精简,并提取标记节点用于分割阴影和遮挡目标;然后利用随机游走的分割结果为Kalman滤波器提供精确的观测信息,以更新滤波器参数.同时,使用基于车底阴影的随机游走算法对目标进行初始分割,以获取Kalman滤波器需要的初始状态向量.实验结果证明,文中算法能够解决运动目标阴影与遮挡问题,并且目标分割平均正确率大于94%,算法满足实时性要求.

An image segmentation algorithm which combines Kalman filter with random walk algorithm is proposed for resolving the problem of shadow and occlusion in traffic video monitoring.Firstly,the predicted information of Kalman filter is used to reduce the working region of the random walk,in which region several mask points are extracted for the segmentation of the shadow and occluded objects.Secondly,the segmentation results of the random walk provide accurate observation information to update the parameters of the Kalman filter.At the same time,a random walk algorithm based on car bottom shadow is proposed to perform initial target segmentation,which is used to obtain the initial state vector of the Kalman filter.Experimental results show that the proposed algorithm can solve the shadow and occlusion problem.And the average accuracy rate of moving vehicle segmentation is more than 94%.Furthermore,the proposed algorithm has real-time performance.